TestController may reuse a view that used window.open(), which prevents process-swapping and causes flakiness

• Component: Tools / Tests
• Status: RESOLVED FIXED

[Yury Semikhatsky]

TestController may reuse Page for which previous test changed Page::openedByDOMWithOpener to true. This may disable PSON for cross-process navigation in subsequent tests and cause flakiness. See https://webkit.org/b/203965#c3 for one example how http/tests/dom/new-window-can-target-opener.html inadvertently disabled PSON for http/tests/inspector/target/target-events-for-provisional-page.html and made it fail.

[Chris Dumez]

trying to find this logic in TestController..

[Chris Dumez]

I think it is more likely we don't swap because: navigation.hasOpenedFrames(). Meaning that the previous test opened another window via window.open().

[Chris Dumez]

Created attachment 383086 [details]
WIP Patch

[Chris Dumez]

Created attachment 383087 [details]
Patch

[Yury Semikhatsky]

(In reply to Chris Dumez from comment #2)
> I think it is more likely we don't swap because:
> navigation.hasOpenedFrames(). Meaning that the previous test opened another
> window via window.open().

FWIW, while debugging the failure I saw

'"Browsing context been opened by DOM without 'noopener'"

coming from 
https://trac.webkit.org/browser/webkit/trunk/Source/WebKit/UIProcess/WebProcessPool.cpp?rev=252014#L2139


Should be easy to reproduce by running:

Tools/Scripts/run-webkit-tests --no-show-results --no-retry-failures --child-processes=1 http/tests/dom/new-window-can-target-opener.html http/tests/inspector/target/target-events-for-provisional-page.html

[Chris Dumez]

(In reply to Yury Semikhatsky from comment #5)
> (In reply to Chris Dumez from comment #2)
> > I think it is more likely we don't swap because:
> > navigation.hasOpenedFrames(). Meaning that the previous test opened another
> > window via window.open().
> 
> FWIW, while debugging the failure I saw
> 
> '"Browsing context been opened by DOM without 'noopener'"
> 
> coming from 
> https://trac.webkit.org/browser/webkit/trunk/Source/WebKit/UIProcess/
> WebProcessPool.cpp?rev=252014#L2139
> 
> 
> Should be easy to reproduce by running:
> 
> Tools/Scripts/run-webkit-tests --no-show-results --no-retry-failures
> --child-processes=1 http/tests/dom/new-window-can-target-opener.html
> http/tests/inspector/target/target-events-for-provisional-page.html

Yes, I have already confirmed that my fix addresses the flakiness with the command you provided.

[Yury Semikhatsky]

(In reply to Chris Dumez from comment #6)
> (In reply to Yury Semikhatsky from comment #5)
> > (In reply to Chris Dumez from comment #2)
> Yes, I have already confirmed that my fix addresses the flakiness with the
> command you provided.

Your change will certainly fix the test failure I was more concerned about the Process/Page being reused in prod. Since you mentioned hasOpenedFrames I wanted to double check if that's actually root cause because it didn't match what I saw in the output. In any case thanks for the quick fix!

[Yury Semikhatsky]

Comment on attachment 383087 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=383087&action=review

lgtm

> Tools/WebKitTestRunner/TestController.cpp:774
> +        if (!m_createdOtherPage && m_mainWebView->viewSupportsOptions(options))

Neat!

[Chris Dumez]

Created attachment 383091 [details]
Patch

[Chris Dumez]

Committed r252228: <https://trac.webkit.org/changeset/252228>

[Radar WebKit Bug Importer]

<rdar://problem/57008023>